| 摘要: |
| 考虑到不同用户任务和计算能力差异，并综合低地球轨道（Low Earth Orbit,LEO）卫星移动和资源限制等因素，针对多颗LEO卫星覆盖场景下的多用户任务决策卸载问题，提出了一种基于深度强化学习（Deep Reinforcement Learning,DRL）的决策卸载和资源分配策略，对系统的耗时和能耗进行优化。决策卸载问题设置为一个离散的有效状态的单用户决策选择问题，采用深度强化学习进行求解。采用拉格朗日乘子法和梯度投影法处理资源分配问题。仿真结果表明，该策略在75次迭代回合后能达到收敛，与其他策略相比，系统成本下降约50%、34%和19%。 |
| 关键词: 低轨卫星 移动边缘计算 资源分配 深度强化学习 决策卸载 |
| DOI：10.20079/j.issn.1001-893x.230506001 |
|
| 基金项目:重庆市自然科学基金面上项目(cstc2021jcyj-msxmX0454) |
|
| A DRL-based Computation Offloading in LEO Satellite MobileEdge Computing System |
| YANG Liming,JIN Yufeng,HUANG Miao |
| (School of Communication and Information Engineering,Chongqing University of Posts and Telecommunications,Chongqing 400065,China) |
| Abstract: |
| In consideration of different user task and computational capacity differences,according to factors such low Earth orbit(LEO) satellite movement and resource constraints,a deep reinforcement learning(DRL) based decision offloading and resource allocation strategy is proposed for the multi-user task decision offloading problem in multiple LEO satellite coverage scenarios to optimize the time and energy consumption of the system.Decision offloading problem is set as a discrete valid state single-user decision selection problem solved by DRL.Resource allocation is handled by the Lagrange multiplier method and the gradient projection method.Simulation results show that the strategy can reach convergence after 75 iteration rounds,and the system cost is reduced by about 50%,34% and 19% compared with other strategies. |
| Key words: LEO satellite mobile edge computing(MEC) resource allocation deep reinforcement learning offloading decision |
